- Email: [email protected]
A Long-term Data Collection System for Life Pattern Sensor
Available online at www.sciencedirect.com
ScienceDirect Procedia Computer Science 22 (2013) 485 – 493
17th International Conference in Knowledge Based and Intelligent Information and Engineering Systems KES2013
A Long-term Data Collection System for Life Pattern Sensor Noriyuki Kushiroa , Taichi Idea , Makoto Katsukurab , Toshiyasu Higumab a Kyushu b Mitsubishi
Institute of Technology, Iizuka, 820-8502, Japan Electric Corporation, Kamakura, 247-8501, Japan
Abstract This paper describes a long-term filed data collection system from a person living house required for designing a life pattern sensor. The life pattern sensor is an activity recognition sensor both for measuring electric power consumption and detecting use of electronic devices by utilizing high-frequency electrical current waveform as a signature of each device. A prototype of the life pattern sensor has been developed and evaluated in five homes. The results show the life pattern sensor measured power consumption within 1% error and identified uses of devices with over 95% accuracy. The capability of the sensor is enough for realizing home energy management system. The long-term data collection system is constructed so as to collect annotated field data for sophisticating the sensor in practical level. In this paper, the overview of the data collecting system and discussion about the following questions based on the field data: 1. Can current waveform be in stable at any conditions in a house? 2. Can sequence of electronic devices describe residents’ real life pattern? 3. Can residents manage energy efficiently by knowing their own life pattern? © 2013 Authors. Published by Elsevier B.V. c 2011 The Published by Elsevier Ltd.. Selection and peer-review under responsibility of KES International Keywords: Life Pattern Sensor, Activity Recognition, Smart Meter, Home Energy Management System
1. Introduction There is a growing interest to introduce renewable energy sources, such as photo-voltaic and wind force electricity systems, for decreasing greenhouse gas emission. Against fluctuation of power generation of them, it is indispensable to realize an energy management system before penetrating renewable energy sources into the market. It is well known that the energy consumption in a home strongly depends on weather and residents’ life pattern. An activity recognition is a key technology for realizing a home energy management system. A life pattern sensor is an activity recognition sensor both for measuring electric power and detecting use of electronic devices by utilizing high-frequency electrical current waveform as a signature of each device. A prototype of the sensor has been developed and installed in five homes prior to this study. We have confirmed that basic capability of the sensor is enough for realizing a home energy management system. For improving technologies of the sensor in practical level, we should answer the following questions: 1. Can current waveform be in stable at any conditions in a house? 2. Can sequence of electronic devices describe residents’ real life pattern? 3. Can residents manage energy efficiently by knowing their own life pattern?
1877-0509 © 2013 The Authors. Published by Elsevier B.V. Selection and peer-review under responsibility of KES International doi:10.1016/j.procs.2013.09.127
486
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
The long-term data collection system is integrated so as to collect annotated field data for sophisticating the sensor in practical level. The overview of the data collecting system and discussion about the above questions are described in the paper. 2. Life Pattern Sensor 2.1. Overview of Life Pattern Sensor A number of approaches to home activity recognition have been considered in recent years. These approaches classified into two categories: direct activity recognition methods with plural of cameras or occupancy sensors installed in a house[1][2], and indirect activity recognition methods with the sequence of objects manipulated by residents[3]. The former methods detects human activities directly with sensors, so that various kinds of residents’ activities can be observed. However, the methods have risks that they cost for installation and intrude residents’ privacy heavily. The later methods have few risks both for cost and privacy intrusion, while the methods have clear limits to what activities can be detected. Low cost and easy-installation is a central focus for energy management systems in residential domain. The methods with a strategically placed sensor on the existing infrastructures (electricity[4], gas[5] and water[6]) gain the spotlight instead of deploying many sensors throughout a house. The Life pattern sensor(Fig.1 ) is a new solution for detecting and classifying the use of electronic devices from a single point of sensing . The sensor relies on the fact that most modern electronics employ switching power supplies to achieve high efficiency. These power supplies continuously generate high frequency electromagnetic interference during operation that propagates throughout power line. The sensor uses small thorn-like interference included in the current waveform as signatures for detecting use of electronic devices(Fig.2).
Figure 1. Prototype of Life Pattern Sensor
Figure 2. Basic Principle of Life Pattern Sensor
2.2. Evaluation of the Life pattern sensor We have installed and evaluated the prototype for a couple of weeks in five houses[7][8]. The results show 95%-100% accuracies for detecting and classifying the use of electronic devices (vacuum cleaner, induction cooker, microwave oven and air conditioner) and within 1% error in measuring power consumption. We have confirmed that the prototype has enough capability for home energy management system.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
487
3. Long-term data collection system 3.1. Data collection system A long-term data collection system is constructed. The system consists of three subsystems: power line monitoring subsystem, use of devices monitoring subsystem and life pattern monitoring subsystemFig.3. 1. The power line monitoring subsystem: the subsystem measures values of power, voltage, current and power factor(P.F.) at every10 seconds and gets snap-shots of current waveforms (20kHz sampling rate) when the power consumption changes over 50W. the subsystem preserves all the parameters of the power line and waveform into the database in Fig.3. 2. The use of devices monitoring subsystem: the subsystem collects ON/OFF status of electronic devices via adapters attached to each device at every 10 seconds and accumulates the history of use of devices. 3. The life pattern monitoring subsystem: the subsystem gathers life events which a resident muttered on Twitter and accumulates the history of the life events. the subsystem also collects status of at-home/out of home with a foot switch operation deployed at entrance. The system has been installed into two houses and has started collecting annotated field data from october in 2012(Fig.4).
Figure 3. Block Diagram of Field Data Collection System
Figure 4. Installation of Field Data Collection System into a House
488
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
3.2. Data analysis tool All data collected by three subsystems are integrated into a database illustrated in Fig.5. The database consists of time stump, status of electronic devices, parameters of power line, life events and current waveforms. A visualization tools has been developed to overview the databaseFig.6). By using the database and the visualization tool, we try to answer the questions described in section.1.
Figure 5. Configuration of the Database of Field Data
Figure 6. Visualization tool for the Database
4. Discussion We have started collecting the annotated field data from October 21, 2012. The following analysis are made on the data from 21/Oct. to 31/Jan. 2013. 4.1. Can current waveform be in stable at any conditions in a house? The waveform transmission from a device to the life pattern sensor is regarded as a kind of power line communication shown in Fig.2. Therefore, the current waveform may be influenced from transmission properties (impedance, attenuation and noise) of domestic and/or external power line in similar to the power line communication. These influences are severe issues to hinder penetration of power line communication for a long time. At first, we ought to discuss the question: can the current waveform be in stable at any conditions in a house? 4.1.1. Stability of power supply The data collection system gathers basic parameters of power line: power, voltage, current and power factor (P.F.) at every 10 seconds. Stability of the power supply is evaluated on the above data. 1. Voltage fluctuation The distribution of voltage is shown in Fig.7. The range, average 103V, maximum 107V and minimum 90V, is almost within the prescribed range by the electric power company. The distribution seems to be the normal distribution.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
2. P.F. fluctuation The distribution of P.F. is shown in Fig.7. The range is that average 0.91, maximum 1.0 and minimum 0.41. The distribution has tree peaks. The results suggest that P.F. is determined by combination of devices used at the same time. 3. Correlation among voltage, power and P.F. The correlation among voltage, power and P.F. is illustrated in Fig.8. The results show that the fluctuation of voltage is independent from P.F. and power. It means that the voltage is influenced only from conditions of power supply, not from conditions of domestic power line and devices. On the other hand, the correspondence between P.F. and power is observed. P.F. tends to become almost 1.0 when power consumption is in high level. From the view point of identifying life events, ad-hoc devices, like a microwave oven, a vacuum cleaner and a hair dryer etc., are significant . Because these devices are only used at the particular life event. Ad-hoc devices generally consume large power. So, P.F. at the time that the life pattern sensor ought to identify the device, is assumed to be almost 1.0. The results lead the conclusion that the current waveform is not influences from impedance of power line. As results, the influences by transmission properties is small for the life pattern sensor design. Probabilistic changes of voltage should be consider in the design.
Figure 7. Distribution of Voltage and Power Factor (P.F)
Figure 8. Correlation among Power, Voltage and Power Factor (P.F)
4.1.2. Influences from load conditions of device 22329 current waveform samples are collected during the test period. The current waveforms are stratified and analyzed independently at each combination of devices with the status of each device field shown in Fig.5. For example, the current waveforms of microwave oven (111samples) are stratified and are illustrated in Fig.9.
489
490
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
Figure 9. Current Waveform (microwave oven)
The results show that the different types of the waveform are mixed. For illustrating features of waveforms in detail, all samples are analyzed with wavelet transform method(Fig.10). In Fig.10, an upper part illustrates a raw data of the waveform, middle and low parts show signal intensity of spectrum from 0Hz to 20KHz. The microwave oven has two types of waveform on its operating conditions. The autocorrelation in each type of waveform is very high. The life sensor should learn two types of waveforms for detecting use of microwave oven. To learn plural of current waveforms for each device impacts memory size and installation cost of the sensor. A breakthrough for deregulating influences from devices’ conditions is a key in the design of the life pattern sensor.
Figure 10. Wavelet Transfer Results of Microwave oven
4.2. Can sequence of electronic devices describe residents’ real life pattern? 4.2.1. Life events and devices Keywords for life events are elicited from mutterings on Twitter and classified in Fig.11. Each life event links to particular devices by using the analysis tool (Fig.6) easily. 4.2.2. Life pattern A life pattern is illustrated by sorting the life events in time sequence (Fig.12). For example, the following life patterns are observed: 1. The resident studied in the laboratory from 10 am to 6 pm on weekdays. 2. The resident often went out on weekend in the first half of the observation period, however, in the latter half of the period, the resident kept studying in the lab. during weekend. 3. The resident washed clothes in the morning twice a week before going to the laboratory. 4. The resident cleaned up the room in simple after returning home on weekdays and cleaned up the room thoroughly at the weekend. 5. The resident usually cooked and ate dinner at home both on weekday and weekend.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
Figure 11. Life events
6. The resident took bath and dried one’s hair in the midnight. 7. The resident often went asleep without turning off devices. 8. The resident studied through the night on Tuesday, and took asleep in the daytime to cancel shortage of sleep on Wednesday. 9. The resident worked at night in the first half of the observation period, however, in the latter half of the period, the resident stopped working and studied in the lab. instead of the part-time job.
Figure 12. Life Pattern (Time Series of Life Events)
As conclusions, some life events can link to particular devices. The sequence of the use of electronic devices can describe residents’ life pattern. On the other hand, the sensor have limits to detect events like sleeping and out-ofhome, at the time when any devices (except normally on device like refrigerator and water heater) are not used.
491
492
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
4.3. Can residents manage energy by knowing their own life pattern? 4.3.1. Relation between life event and power consumption The power consumption and duration time required for each life event are defined based on the database. The results are shown in Table.1. Table 1. Relation between Life Event and Power Consumption and Duration of Time
Life events Sleeping Sleeping(without turnoff) Studying Dining Cleaning(corner-cutting) Cleaning(throughly) Washing Cocking(ultra-simple) Cocking(simple) Cocking(full) Part-time job Sports Bath Bath preparation Drying hair Out of home Heating Drying Other
Power(W) 413+-217 813+-273 344+-13 1561+-444 4009+-0 1113+-137 1640+-177 1536+-288 4339+-1128 3758+-1115 383+-64 336+-6 1502+-383 1688+-371 6605+-1183 365+-90 1133+-359 292+-0 1159+-437
Duration(minutes) 413+-138 549+-167 535+-257 32+-23 5+-0 104+-7 38+-6 5+-1 29+-18 31+-27 203+-143 185+-57 23+-28 54+-38 2+-1 129+-240 62+-99 298+-1 49+-88
4.3.2. Life pattern based energy control A typical life pattern of weekday is defined on the life pattern shown in Fig.13. The demand of power in a day is predicted with the series of life events and the amount of power consumed for each life event defined in Table.1. The result of power consumption, when the resident lived on the typical life pattern actually, is also illustrated in Fig.13.
Figure 13. Energy Control based on Life Pattern
We confirm that the demand of power in a day is predicted with the series of life events and the amount of power consumed for each life events. It is suggested that energy management might be possible by knowing the life pattern.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
493
5. Conclusion The long-term data collection system is constructed so as to answer the following questions: 1. Can current waveform be in stable at any conditions in a house? 2. Can sequence of electronic devices describe residents’ real life pattern? 3. Can residents manage home energy efficiently by knowing their own life pattern? The system has been installed into two houses and has started collecting field data from october in 2012. The following analysis are made on the data from 21/Oct. to 31/Jan. 2013. 1. Stability of the waveform: The waveform is stable for the characteristics of domestic power line. We should focus on the following two factors in the design of the life pattern sensor: the fluctuation of voltage and the waveform variations depend on device’s operating mode. 2. Relation between life event and device: Some life events link to particular devices. Sequence of electronic devices describes residents’ real life pattern. On the other hand, the sensor have limits to detect life events like sleeping and out of home at the time when any devices are used. 3. Capability of energy management by knowing the life pattern: The demand of power in a day is predicted with the series of life events and the amount of power consumed for each life events. We are going to collect field data for two years. Based on the field data, we improve an algorithmic of the life pattern sensor and implement as embedded software for a smart meter to promote the sensor in practical use. References [1] Shoaib et al.: Context-aware visual analysis of elderly activity in a cluttered home environment. EURASIP Journal on Advances in Signal Processing 2011 [2] Nait-Charif, H.; McKenna, S.J., ”Activity summarisation and fall detection in a supportive home environment,” Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on , vol.4, no., pp.323,326 Vol.4, pp.23-26 Aug. (2004) [3] Jianxin Wu; Osuntogun, A.; Choudhury, T.; Philipose, M.; Rehg, J.M., ”A Scalable Approach to Activity Recognition based on Object Use,” Computer Vision, 2007. ICCV 2007. IEEE 11th International Conference on , vol., no., pp.1,8, 14-21 Oct. (2007) [4] Sidhant Gupta, Matthew S. Reynolds, Shwetak N. Patel: ElectriSense: single-point sensing using EMI for electrical event detection and classification in the home. UbiComp 2010: pp.139-148, (2010) [5] James Fogarty, Carolyn Au, and Scott E. Hudson. 2006. Sensing from the basement: a feasibility study of unobtrusive and low-cost home activity recognition. In Proceedings of the 19th annual ACM symposium on User interface software and technology. ACM, pp.91-100, (2006) [6] Gabe Cohn, Sidhant Gupta, Jon Froehlich, Eric Larson, and Shwetak N. Patel. 2010. GasSense: appliance-level, single-point sensing of gas activity in the home. In Proceedings of the 8th international conference on Pervasive Computing,Springer-Verlag, pp.265-282,(2010) [7] Makoto Katsukura, Masahiro Nakata, Yoshiaki Ito and Noriyuki Kushiro: Life Pattern Sensor with Non-intrusive Appliance Monitoring, IEEE International Conference on Consumer Electronics, 2009 [8] Noriyuki Kushiro, Makoto Katsukura, Masanori Nakata and Yoshiaki Ito: Non-intrusive Human Behavior Sensor for Health Care System, Human Computer Interaction International Conference, (2009)
ScienceDirect Procedia Computer Science 22 (2013) 485 – 493
17th International Conference in Knowledge Based and Intelligent Information and Engineering Systems KES2013
A Long-term Data Collection System for Life Pattern Sensor Noriyuki Kushiroa , Taichi Idea , Makoto Katsukurab , Toshiyasu Higumab a Kyushu b Mitsubishi
Institute of Technology, Iizuka, 820-8502, Japan Electric Corporation, Kamakura, 247-8501, Japan
Abstract This paper describes a long-term filed data collection system from a person living house required for designing a life pattern sensor. The life pattern sensor is an activity recognition sensor both for measuring electric power consumption and detecting use of electronic devices by utilizing high-frequency electrical current waveform as a signature of each device. A prototype of the life pattern sensor has been developed and evaluated in five homes. The results show the life pattern sensor measured power consumption within 1% error and identified uses of devices with over 95% accuracy. The capability of the sensor is enough for realizing home energy management system. The long-term data collection system is constructed so as to collect annotated field data for sophisticating the sensor in practical level. In this paper, the overview of the data collecting system and discussion about the following questions based on the field data: 1. Can current waveform be in stable at any conditions in a house? 2. Can sequence of electronic devices describe residents’ real life pattern? 3. Can residents manage energy efficiently by knowing their own life pattern? © 2013 Authors. Published by Elsevier B.V. c 2011 The Published by Elsevier Ltd.. Selection and peer-review under responsibility of KES International Keywords: Life Pattern Sensor, Activity Recognition, Smart Meter, Home Energy Management System
1. Introduction There is a growing interest to introduce renewable energy sources, such as photo-voltaic and wind force electricity systems, for decreasing greenhouse gas emission. Against fluctuation of power generation of them, it is indispensable to realize an energy management system before penetrating renewable energy sources into the market. It is well known that the energy consumption in a home strongly depends on weather and residents’ life pattern. An activity recognition is a key technology for realizing a home energy management system. A life pattern sensor is an activity recognition sensor both for measuring electric power and detecting use of electronic devices by utilizing high-frequency electrical current waveform as a signature of each device. A prototype of the sensor has been developed and installed in five homes prior to this study. We have confirmed that basic capability of the sensor is enough for realizing a home energy management system. For improving technologies of the sensor in practical level, we should answer the following questions: 1. Can current waveform be in stable at any conditions in a house? 2. Can sequence of electronic devices describe residents’ real life pattern? 3. Can residents manage energy efficiently by knowing their own life pattern?
1877-0509 © 2013 The Authors. Published by Elsevier B.V. Selection and peer-review under responsibility of KES International doi:10.1016/j.procs.2013.09.127
486
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
The long-term data collection system is integrated so as to collect annotated field data for sophisticating the sensor in practical level. The overview of the data collecting system and discussion about the above questions are described in the paper. 2. Life Pattern Sensor 2.1. Overview of Life Pattern Sensor A number of approaches to home activity recognition have been considered in recent years. These approaches classified into two categories: direct activity recognition methods with plural of cameras or occupancy sensors installed in a house[1][2], and indirect activity recognition methods with the sequence of objects manipulated by residents[3]. The former methods detects human activities directly with sensors, so that various kinds of residents’ activities can be observed. However, the methods have risks that they cost for installation and intrude residents’ privacy heavily. The later methods have few risks both for cost and privacy intrusion, while the methods have clear limits to what activities can be detected. Low cost and easy-installation is a central focus for energy management systems in residential domain. The methods with a strategically placed sensor on the existing infrastructures (electricity[4], gas[5] and water[6]) gain the spotlight instead of deploying many sensors throughout a house. The Life pattern sensor(Fig.1 ) is a new solution for detecting and classifying the use of electronic devices from a single point of sensing . The sensor relies on the fact that most modern electronics employ switching power supplies to achieve high efficiency. These power supplies continuously generate high frequency electromagnetic interference during operation that propagates throughout power line. The sensor uses small thorn-like interference included in the current waveform as signatures for detecting use of electronic devices(Fig.2).
Figure 1. Prototype of Life Pattern Sensor
Figure 2. Basic Principle of Life Pattern Sensor
2.2. Evaluation of the Life pattern sensor We have installed and evaluated the prototype for a couple of weeks in five houses[7][8]. The results show 95%-100% accuracies for detecting and classifying the use of electronic devices (vacuum cleaner, induction cooker, microwave oven and air conditioner) and within 1% error in measuring power consumption. We have confirmed that the prototype has enough capability for home energy management system.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
487
3. Long-term data collection system 3.1. Data collection system A long-term data collection system is constructed. The system consists of three subsystems: power line monitoring subsystem, use of devices monitoring subsystem and life pattern monitoring subsystemFig.3. 1. The power line monitoring subsystem: the subsystem measures values of power, voltage, current and power factor(P.F.) at every10 seconds and gets snap-shots of current waveforms (20kHz sampling rate) when the power consumption changes over 50W. the subsystem preserves all the parameters of the power line and waveform into the database in Fig.3. 2. The use of devices monitoring subsystem: the subsystem collects ON/OFF status of electronic devices via adapters attached to each device at every 10 seconds and accumulates the history of use of devices. 3. The life pattern monitoring subsystem: the subsystem gathers life events which a resident muttered on Twitter and accumulates the history of the life events. the subsystem also collects status of at-home/out of home with a foot switch operation deployed at entrance. The system has been installed into two houses and has started collecting annotated field data from october in 2012(Fig.4).
Figure 3. Block Diagram of Field Data Collection System
Figure 4. Installation of Field Data Collection System into a House
488
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
3.2. Data analysis tool All data collected by three subsystems are integrated into a database illustrated in Fig.5. The database consists of time stump, status of electronic devices, parameters of power line, life events and current waveforms. A visualization tools has been developed to overview the databaseFig.6). By using the database and the visualization tool, we try to answer the questions described in section.1.
Figure 5. Configuration of the Database of Field Data
Figure 6. Visualization tool for the Database
4. Discussion We have started collecting the annotated field data from October 21, 2012. The following analysis are made on the data from 21/Oct. to 31/Jan. 2013. 4.1. Can current waveform be in stable at any conditions in a house? The waveform transmission from a device to the life pattern sensor is regarded as a kind of power line communication shown in Fig.2. Therefore, the current waveform may be influenced from transmission properties (impedance, attenuation and noise) of domestic and/or external power line in similar to the power line communication. These influences are severe issues to hinder penetration of power line communication for a long time. At first, we ought to discuss the question: can the current waveform be in stable at any conditions in a house? 4.1.1. Stability of power supply The data collection system gathers basic parameters of power line: power, voltage, current and power factor (P.F.) at every 10 seconds. Stability of the power supply is evaluated on the above data. 1. Voltage fluctuation The distribution of voltage is shown in Fig.7. The range, average 103V, maximum 107V and minimum 90V, is almost within the prescribed range by the electric power company. The distribution seems to be the normal distribution.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
2. P.F. fluctuation The distribution of P.F. is shown in Fig.7. The range is that average 0.91, maximum 1.0 and minimum 0.41. The distribution has tree peaks. The results suggest that P.F. is determined by combination of devices used at the same time. 3. Correlation among voltage, power and P.F. The correlation among voltage, power and P.F. is illustrated in Fig.8. The results show that the fluctuation of voltage is independent from P.F. and power. It means that the voltage is influenced only from conditions of power supply, not from conditions of domestic power line and devices. On the other hand, the correspondence between P.F. and power is observed. P.F. tends to become almost 1.0 when power consumption is in high level. From the view point of identifying life events, ad-hoc devices, like a microwave oven, a vacuum cleaner and a hair dryer etc., are significant . Because these devices are only used at the particular life event. Ad-hoc devices generally consume large power. So, P.F. at the time that the life pattern sensor ought to identify the device, is assumed to be almost 1.0. The results lead the conclusion that the current waveform is not influences from impedance of power line. As results, the influences by transmission properties is small for the life pattern sensor design. Probabilistic changes of voltage should be consider in the design.
Figure 7. Distribution of Voltage and Power Factor (P.F)
Figure 8. Correlation among Power, Voltage and Power Factor (P.F)
4.1.2. Influences from load conditions of device 22329 current waveform samples are collected during the test period. The current waveforms are stratified and analyzed independently at each combination of devices with the status of each device field shown in Fig.5. For example, the current waveforms of microwave oven (111samples) are stratified and are illustrated in Fig.9.
489
490
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
Figure 9. Current Waveform (microwave oven)
The results show that the different types of the waveform are mixed. For illustrating features of waveforms in detail, all samples are analyzed with wavelet transform method(Fig.10). In Fig.10, an upper part illustrates a raw data of the waveform, middle and low parts show signal intensity of spectrum from 0Hz to 20KHz. The microwave oven has two types of waveform on its operating conditions. The autocorrelation in each type of waveform is very high. The life sensor should learn two types of waveforms for detecting use of microwave oven. To learn plural of current waveforms for each device impacts memory size and installation cost of the sensor. A breakthrough for deregulating influences from devices’ conditions is a key in the design of the life pattern sensor.
Figure 10. Wavelet Transfer Results of Microwave oven
4.2. Can sequence of electronic devices describe residents’ real life pattern? 4.2.1. Life events and devices Keywords for life events are elicited from mutterings on Twitter and classified in Fig.11. Each life event links to particular devices by using the analysis tool (Fig.6) easily. 4.2.2. Life pattern A life pattern is illustrated by sorting the life events in time sequence (Fig.12). For example, the following life patterns are observed: 1. The resident studied in the laboratory from 10 am to 6 pm on weekdays. 2. The resident often went out on weekend in the first half of the observation period, however, in the latter half of the period, the resident kept studying in the lab. during weekend. 3. The resident washed clothes in the morning twice a week before going to the laboratory. 4. The resident cleaned up the room in simple after returning home on weekdays and cleaned up the room thoroughly at the weekend. 5. The resident usually cooked and ate dinner at home both on weekday and weekend.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
Figure 11. Life events
6. The resident took bath and dried one’s hair in the midnight. 7. The resident often went asleep without turning off devices. 8. The resident studied through the night on Tuesday, and took asleep in the daytime to cancel shortage of sleep on Wednesday. 9. The resident worked at night in the first half of the observation period, however, in the latter half of the period, the resident stopped working and studied in the lab. instead of the part-time job.
Figure 12. Life Pattern (Time Series of Life Events)
As conclusions, some life events can link to particular devices. The sequence of the use of electronic devices can describe residents’ life pattern. On the other hand, the sensor have limits to detect events like sleeping and out-ofhome, at the time when any devices (except normally on device like refrigerator and water heater) are not used.
491
492
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
4.3. Can residents manage energy by knowing their own life pattern? 4.3.1. Relation between life event and power consumption The power consumption and duration time required for each life event are defined based on the database. The results are shown in Table.1. Table 1. Relation between Life Event and Power Consumption and Duration of Time
Life events Sleeping Sleeping(without turnoff) Studying Dining Cleaning(corner-cutting) Cleaning(throughly) Washing Cocking(ultra-simple) Cocking(simple) Cocking(full) Part-time job Sports Bath Bath preparation Drying hair Out of home Heating Drying Other
Power(W) 413+-217 813+-273 344+-13 1561+-444 4009+-0 1113+-137 1640+-177 1536+-288 4339+-1128 3758+-1115 383+-64 336+-6 1502+-383 1688+-371 6605+-1183 365+-90 1133+-359 292+-0 1159+-437
Duration(minutes) 413+-138 549+-167 535+-257 32+-23 5+-0 104+-7 38+-6 5+-1 29+-18 31+-27 203+-143 185+-57 23+-28 54+-38 2+-1 129+-240 62+-99 298+-1 49+-88
4.3.2. Life pattern based energy control A typical life pattern of weekday is defined on the life pattern shown in Fig.13. The demand of power in a day is predicted with the series of life events and the amount of power consumed for each life event defined in Table.1. The result of power consumption, when the resident lived on the typical life pattern actually, is also illustrated in Fig.13.
Figure 13. Energy Control based on Life Pattern
We confirm that the demand of power in a day is predicted with the series of life events and the amount of power consumed for each life events. It is suggested that energy management might be possible by knowing the life pattern.
Noriyuki Kushiro et al. / Procedia Computer Science 22 (2013) 485 – 493
493
5. Conclusion The long-term data collection system is constructed so as to answer the following questions: 1. Can current waveform be in stable at any conditions in a house? 2. Can sequence of electronic devices describe residents’ real life pattern? 3. Can residents manage home energy efficiently by knowing their own life pattern? The system has been installed into two houses and has started collecting field data from october in 2012. The following analysis are made on the data from 21/Oct. to 31/Jan. 2013. 1. Stability of the waveform: The waveform is stable for the characteristics of domestic power line. We should focus on the following two factors in the design of the life pattern sensor: the fluctuation of voltage and the waveform variations depend on device’s operating mode. 2. Relation between life event and device: Some life events link to particular devices. Sequence of electronic devices describes residents’ real life pattern. On the other hand, the sensor have limits to detect life events like sleeping and out of home at the time when any devices are used. 3. Capability of energy management by knowing the life pattern: The demand of power in a day is predicted with the series of life events and the amount of power consumed for each life events. We are going to collect field data for two years. Based on the field data, we improve an algorithmic of the life pattern sensor and implement as embedded software for a smart meter to promote the sensor in practical use. References [1] Shoaib et al.: Context-aware visual analysis of elderly activity in a cluttered home environment. EURASIP Journal on Advances in Signal Processing 2011 [2] Nait-Charif, H.; McKenna, S.J., ”Activity summarisation and fall detection in a supportive home environment,” Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on , vol.4, no., pp.323,326 Vol.4, pp.23-26 Aug. (2004) [3] Jianxin Wu; Osuntogun, A.; Choudhury, T.; Philipose, M.; Rehg, J.M., ”A Scalable Approach to Activity Recognition based on Object Use,” Computer Vision, 2007. ICCV 2007. IEEE 11th International Conference on , vol., no., pp.1,8, 14-21 Oct. (2007) [4] Sidhant Gupta, Matthew S. Reynolds, Shwetak N. Patel: ElectriSense: single-point sensing using EMI for electrical event detection and classification in the home. UbiComp 2010: pp.139-148, (2010) [5] James Fogarty, Carolyn Au, and Scott E. Hudson. 2006. Sensing from the basement: a feasibility study of unobtrusive and low-cost home activity recognition. In Proceedings of the 19th annual ACM symposium on User interface software and technology. ACM, pp.91-100, (2006) [6] Gabe Cohn, Sidhant Gupta, Jon Froehlich, Eric Larson, and Shwetak N. Patel. 2010. GasSense: appliance-level, single-point sensing of gas activity in the home. In Proceedings of the 8th international conference on Pervasive Computing,Springer-Verlag, pp.265-282,(2010) [7] Makoto Katsukura, Masahiro Nakata, Yoshiaki Ito and Noriyuki Kushiro: Life Pattern Sensor with Non-intrusive Appliance Monitoring, IEEE International Conference on Consumer Electronics, 2009 [8] Noriyuki Kushiro, Makoto Katsukura, Masanori Nakata and Yoshiaki Ito: Non-intrusive Human Behavior Sensor for Health Care System, Human Computer Interaction International Conference, (2009)